Function score

evaluation/eval_length.py:5–9

(x, y)

import numpy as np

def score(x, y):
    if y > x:
        return 100 * max(0, 1. - (y / x - 1) / 3)
    else:
        return 100 * max(0, 1. - (x / y - 1) / 2)

model = "LongWriter-glm4-9b"
prediction = [json.loads(line) for line in open(f'models/{model}/pred.jsonl', encoding='utf-8')]